以内蒙古旺业甸林场为研究区,结合地面调查,对高分二号遥感数据进行预处理,并提取光谱信息、植被指数及纹理信息等48个遥感因子,采用Pearson相关系数法筛选出8个因子进行建模。采用多元线性回归、多层感知机、K-近邻、支持向量机、随机森林模型估测森林蓄积量,得到研究区内森林蓄积量反演图。结果表明:1)高分二号影像提取的遥感因子中,基于二阶矩阵的纹理特征均值(Mean)与森林蓄积量的相关性较高;2)随机森林相对于多元线性、多层感知机、K-近邻、支持向量机等方法具有更好的森林蓄积量估测精度,其相对均方根误差(rRMSE)为25.40%;3)研究区内森林蓄积量高的地区主要分布在西部和东南部;森林蓄积量低的地区主要分布在西北部、中部及北部,与实际调查情况一致。国产高分二号影像利用随机森林算法在森林蓄积量反演方面具有一定的潜力。

Taking the Wangyedian Forest Farm in Inner Mongolia as the research area,combined with ground surveys,and based on the preprocessing of the GF-2 remote sensing data,48 remote sensing factors such as spectral information,vegetation index and texture information were extracted,and 8 remote sensing factors were selected for modeling by Pearson correlation coefficient method.Using multiple linear regression,multi-layer perceptron,K-nearest neighbor,support vector machine,and random forest model to estimate the forest volume,the forest volume inversion map in the study area was obtained.The results showed that:1) Among the remote sensing factors extracted from GF-2,mean of texture features based on the second-order matrix had a higher correlation with the forest volume;2) Random Forest had better estimation accuracy of forest volume than methods such as multiple linear regression,multi-layer perceptron,K-nearest neighbor and support vector machine,and its relative root mean square error (rRMSE) was 25.40%;3) The areas with high forest volume in the study area were mainly distributed in the west and southeast;the areas with low forest volume were mainly distributed in the northwest,central and northern parts,which were consistent with the actual investigation.The domestic GF-2 image and random forest algorithm had certain potential in the inversion of forest volume.